This Billionaire Just Loaded Up on 5 Genius AI Picks

Appaloosa (David Tepper) materially increased Q4 positions: Alphabet +29%, Meta +62%, Microsoft +8%; Micron holdings rose 200% and TSMC was increased 7% (TSMC now ~5% of the portfolio). Since Q4 highs these names have pulled back (Meta down ~20% since start of Q4 2025; Microsoft down ~30%), presenting what the author calls bargain opportunities as the companies spend heavily on AI infrastructure. Tepper is positioning across AI hyperscalers and chipmakers as secular AI demand grows; TSMC forecasts AI chip revenue CAGR in the mid- to high-50% range for 2024–2029 and Micron projects HBM demand rising from $35B in 2025 to $100B by 2028.

Analysis

The dominant trade here is a structural build-out of AI compute capacity, which benefits scale-exposed component suppliers (memory and foundry ecosystems) more reliably than single-GPU winners. Expect pricing power concentrated where capacity is scarce and switching costs are high: HBM/memory stacks and leading-node foundries will see multi-quarter order stickiness, while commodity DRAM and trailing-node fabs will experience sharper cyclicality.

Second-order supply effects matter: packaging, test & assembly, and substrate vendors will be the choke points that determine who can actually monetize wafer allocations. That creates asymmetric opportunities — firms that own advanced packaging capacity (or long-term substrate contracts) will capture outsized margin expansion even if wafer ASPs compress later.

Key risks are timing and demand elasticity. If hyperscaler monetization of AI workloads lags, capex plans can be deferred quickly and create a 6–18 month inventory hangover across memory and wafer suppliers. Conversely, an acceleration in generative AI SaaS monetization would compress ROIC payback below current market expectations and re-rate both memory and foundry names.

The positioning imbalance is also noteworthy: passive/quant flows have concentrated into a handful of hyperscaler names, amplifying downside on sentiment shifts but muting upside for component suppliers where direct fundamentals drive earnings revisions. That suggests asymmetric trade construction where long component exposure plus small hedges against hyperscaler idiosyncratic risk offers favorable risk/reward.

Key Decisions for Investors

• Long MU (core equity or 9–12 month LEAPs) — Size 3–5% portfolio. Rationale: direct exposure to HBM/DRAM tightness and favorable long-cycle secular demand. Target 40–70% upside over 12–24 months; hard stop or hedge if shares rise 30% intra-trade (lock profits) or drop 25% (buy protective puts).
• Long TSM (shares + short 6–9 month 5–10% OTM calls) — Size 3–5%. Rationale: foundry pricing power and backlog stickiness from advanced-node tooling and packaging. Target 30–50% total return in 12–36 months while collecting premium; exit or reduce if backlog transparency shows sequential decline for two quarters.
• Paired trade: Long MU + TSM vs small hedge via 3–6 month put spreads on GOOG/MSFT/META (equal-weighted) — Net risk: pay small premium to protect against sudden capex pulls by hyperscalers. Use put spreads to cap hedge cost; ideal if hyperscaler sentiment reverses quickly, preserving component upside while limiting drawdown to 6–8% portfolio-equivalent.
• Contrarian short idea: Selective INTC short via 6–12 month puts (size 1–2%) — Thesis: if foundry share gains remain concentrated with leading-node fabs, legacy integrated CPU vendors face margin compression and capital intensity that delays ROIC recovery. Risk: execution or product-cycle surprises at Intel — cap position size accordingly.